Quickstart
Create a workspace API key and get your first cited answer from /v1/query in under a minute.
Braize answers questions from your own knowledge base, with a citation on every factual claim. This page gets you from a fresh account to a real, grounded answer over the REST API.
1. Create a workspace and add a source
Sign in, create a workspace (pick the region you want your data pinned to — this cannot be changed later without a migration), and add at least one source: upload a file or connect a website page. You can also click Try it on sample data to skip straight to step 3 while your own documents index.
2. Mint an API key
In the dashboard, go to Settings → API keys and create a key. Keys are
prefixed bzk_ so they're recognizable in logs and diffs — treat the full
value as a secret; it is shown once. Export it in your shell:
export BRAIZE_API_KEY="bzk_..."3. Ask your first question
curl -s https://apibr.aize.dev/v1/query \
-H "Authorization: Bearer $BRAIZE_API_KEY" \
-H "Content-Type: application/json" \
-d '{"text": "What does this workspace know about?"}'Every response is wrapped in an envelope: {"data": {...}, "meta": {...}}.
data.blocks carries the answer's typed content blocks — citation blocks by
default, plus a generated prose block if your workspace has
include_generated_answers turned on. data.confidence is a 0–1 score; when
Braize doesn't have grounding for your question, it says so rather than
guessing.
4. Or retrieve passages instead of a generated answer
If you're building your own agent and want the ranked source passages
without generation latency, call POST /v1/retrieve instead — same
permission-aware search, no generation cost. Walkthrough:
Agent-native retrieval (full request/response shape in the
API reference).
What's next
- Authentication & regions — how API keys, regions,
and the
Braize-Versionheader work. - Asking questions & streaming — every
/v1/queryfield, plus the SSE event schema. - Ingesting documents — upload, URL, and source ingestion beyond what you added during setup.
- Rate limits & quotas — the free-beta caps and what a 429 looks like.
- Handling errors — the RFC 9457 Problem shape every error uses.
- Setting up MCP — call Braize from Claude, Cursor, or any MCP-capable agent.
- Core concepts — why citations, ACL pre-filtering, and freshness are the product, not add-ons.
- API & MCP reference — every endpoint and tool, generated straight from the contract.